From e35337f05eefc003b47fe83b60601e37010c4f21 Mon Sep 17 00:00:00 2001
From: Tim Dettmers <tim.dettmers@gmail.com>
Date: Fri, 5 Aug 2022 07:13:24 -0700
Subject: Now determining cuda version via libcudart.so call.

---
 bitsandbytes/cuda_setup/main.py  | 69 ++++++++++++++++++++++++++--------------
 bitsandbytes/cuda_setup/paths.py |  2 +-
 2 files changed, 47 insertions(+), 24 deletions(-)

(limited to 'bitsandbytes/cuda_setup')

diff --git a/bitsandbytes/cuda_setup/main.py b/bitsandbytes/cuda_setup/main.py
index 1e52f89..f1c845c 100644
--- a/bitsandbytes/cuda_setup/main.py
+++ b/bitsandbytes/cuda_setup/main.py
@@ -28,10 +28,40 @@ def check_cuda_result(cuda, result_val):
     if result_val != 0:
         error_str = ctypes.c_char_p()
         cuda.cuGetErrorString(result_val, ctypes.byref(error_str))
-        raise Exception(f"CUDA exception! ERROR: {error_str}")
+        raise Exception(f"CUDA exception! Error code: {error_str.value.decode()}")
 
+def get_cuda_version(cuda, cudart_path):
+    # https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART____VERSION.html#group__CUDART____VERSION
+    try:
+        cudart = ctypes.CDLL(cudart_path)
+    except OSError:
+        # TODO: shouldn't we error or at least warn here?
+        print(f'ERROR: libcudart.so could not be read from path: {cudart_path}!')
+        return None
+
+    version = ctypes.c_int()
+    check_cuda_result(cuda, cudart.cudaRuntimeGetVersion(ctypes.byref(version)))
+    version = int(version.value)
+    major = version//1000
+    minor = (version-(major*1000))//10
+
+    return f'{major}{minor}'
+
+
+def get_cuda_lib_handle():
+    # 1. find libcuda.so library (GPU driver) (/usr/lib)
+    try:
+        cuda = ctypes.CDLL("libcuda.so")
+    except OSError:
+        # TODO: shouldn't we error or at least warn here?
+        print('ERROR: libcuda.so not found!')
+        return None
+    check_cuda_result(cuda, cuda.cuInit(0))
 
-def get_compute_capabilities():
+    return cuda
+
+
+def get_compute_capabilities(cuda):
     """
     1. find libcuda.so library (GPU driver) (/usr/lib)
        init_device -> init variables -> call function by reference
@@ -42,13 +72,6 @@ def get_compute_capabilities():
     # bits taken from https://gist.github.com/f0k/63a664160d016a491b2cbea15913d549
     """
 
-    # 1. find libcuda.so library (GPU driver) (/usr/lib)
-    try:
-        cuda = ctypes.CDLL("libcuda.so")
-    except OSError:
-        # TODO: shouldn't we error or at least warn here?
-        print('ERROR: libcuda.so not found!')
-        return None
 
     nGpus = ctypes.c_int()
     cc_major = ctypes.c_int()
@@ -57,8 +80,6 @@ def get_compute_capabilities():
     result = ctypes.c_int()
     device = ctypes.c_int()
 
-    check_cuda_result(cuda, cuda.cuInit(0))
-
     check_cuda_result(cuda, cuda.cuDeviceGetCount(ctypes.byref(nGpus)))
     ccs = []
     for i in range(nGpus.value):
@@ -75,13 +96,13 @@ def get_compute_capabilities():
 
 
 # def get_compute_capability()-> Union[List[str, ...], None]: # FIXME: error
-def get_compute_capability():
+def get_compute_capability(cuda):
     """
     Extracts the highest compute capbility from all available GPUs, as compute
     capabilities are downwards compatible. If no GPUs are detected, it returns
     None.
     """
-    ccs = get_compute_capabilities()
+    ccs = get_compute_capabilities(cuda)
     if ccs is not None:
         # TODO: handle different compute capabilities; for now, take the max
         return ccs[-1]
@@ -89,10 +110,19 @@ def get_compute_capability():
 
 
 def evaluate_cuda_setup():
-    cuda_path = determine_cuda_runtime_lib_path()
-    print(f"CUDA SETUP: CUDA path found: {cuda_path}")
-    cc = get_compute_capability()
     binary_name = "libbitsandbytes_cpu.so"
+    cudart_path = determine_cuda_runtime_lib_path()
+    if cudart_path is None:
+        print(
+            "WARNING: No libcudart.so found! Install CUDA or the cudatoolkit package (anaconda)!"
+        )
+        return binary_name
+
+    print(f"CUDA SETUP: CUDA path found: {cudart_path}")
+    cuda = get_cuda_lib_handle()
+    cc = get_compute_capability(cuda)
+    cuda_version_string = get_cuda_version(cuda, cudart_path)
+
 
     if cc == '':
         print(
@@ -107,15 +137,8 @@ def evaluate_cuda_setup():
     # (1) CUDA missing cases (no CUDA installed by CUDA driver (nvidia-smi accessible)
     # (2) Multiple CUDA versions installed
 
-    # FIXME: cuda_home is still unused
-    cuda_home = str(Path(cuda_path).parent.parent)
     # we use ls -l instead of nvcc to determine the cuda version
     # since most installations will have the libcudart.so installed, but not the compiler
-    ls_output, err = execute_and_return(f"ls -l {cuda_path}")
-    major, minor, revision = (
-        ls_output.split(" ")[-1].replace("libcudart.so.", "").split(".")
-    )
-    cuda_version_string = f"{major}{minor}"
     print(f'CUDA_SETUP: Detected CUDA version {cuda_version_string}')
 
     def get_binary_name():
diff --git a/bitsandbytes/cuda_setup/paths.py b/bitsandbytes/cuda_setup/paths.py
index c4a7465..b5e04f8 100644
--- a/bitsandbytes/cuda_setup/paths.py
+++ b/bitsandbytes/cuda_setup/paths.py
@@ -123,4 +123,4 @@ def determine_cuda_runtime_lib_path() -> Union[Path, None]:
 
     warn_in_case_of_duplicates(cuda_runtime_libs)
 
-    return next(iter(cuda_runtime_libs)) if cuda_runtime_libs else set()
+    return next(iter(cuda_runtime_libs)) if cuda_runtime_libs else None
-- 
cgit v1.2.3